A red humanoid research robot standing against a black backdrop

vizmaya desk · 4 November 2026

Will androids outnumber humans?

A new physics paper counts the robots and counts the people. The answer hangs on one question: who builds the next robot?

How fast would robots have to grow?

Photo: Cadop / Wikimedia Commons, CC BY-SA 4.0. Paper: arXiv:2611.03242

The number to beat

73.5%

That is the growth robots need every year, for 25 years, to match the number of people by 2050.

They start from about 10,000 humanoid robots in 2025.

Is anyone actually predicting that?

Source: paper p.25. Computed: ln(9.664e9 / 1e4) / 25 = 0.551 a year, then e^0.551 - 1 = 0.735.

One widely quoted forecast: about 1 billion humanoid robots by 2050.

That is Morgan Stanley's estimate. It expects slow growth until at least 2035, then a fast climb.

CONTESTED: one bank's forecast

Is 1 billion enough to match us?

Sources: paper p.10; Morgan Stanley, humanoid.guide

Atlas, a two-legged humanoid research robot, seen from the front
Atlas, a research humanoid. Photo: DARPA / Wikimedia Commons, public domain.

People for every humanoid robot, today

823,000

That is 8.23 billion people divided by an estimated 10,000 humanoid robots in 2025.

CONTESTED: no one has a real robot census

Does the 2050 forecast close that gap?

Sources: paper p.8, p.10; Chamie (IPS, 2025). Computed: 8.23e9 / 1e4 = 823,000.

Ameca, a humanoid robot with a grey face and exposed metal arms and torso
Ameca, a humanoid robot. Photo: Willy Jackson / Wikimedia Commons, CC BY-SA 4.0.

By 2050 the gap shrinks to 9.66 people per robot. Still not a tie.

People grow slowly, from 8.23 to 9.66 billion. Robots leap from 10,000 to 1 billion.

SCENARIO: the robot line is a forecast, not data

Robots grow so fast. Why do they still lose?

Sources: paper Fig. 1 p.12, p.8, p.10. Robot points computed from the US timing shape, p.10.

In 2050, robots grow about 20 times faster than people.

People grow 0.49% a year. Robots grow 9.8% a year. Whoever grows faster gains share. But a fast percent on a small number is still a small number.

So what if we keep going to 2100?

Sources: paper p.9 (humans), p.11 and p.23 (robots), Eq. 37 p.15. Computed: 9.8 / 0.49 is about 20.

By 2100 robots still trail: 3.96 billion vs 10.19 billion.

This is what happens if the factory-built growth pattern simply keeps running. From 2050 to 2100 robots average about 2.79% growth a year. They reach 0.388 of the human count. No tie.

What would change that?

Source: paper p.23. Computed: 3.96 / 10.19 = 0.388.

Factories add robots. Robots could add robots too.

Today people build robots in factories. Add 100 a year, and the pile grows in a straight line.

If each robot also helps build new ones, the pile snowballs. After 20 years: about 2,100 robots vs about 32,300.

ILLUSTRATION: made-up round numbers

When does a robot make more than it loses?

Idea: paper p.3-4, p.17-18. Illustration: start 100; factory +100 a year; robots add 0.2 each a year.

One number decides if robots can build themselves.

Call it P. It is what each robot helps build in a year, divided by the share of robots that wear out in a year.

Above 1, the pile grows on its own. Below 1, it fades. In the paper's 4,096-run test, a tie by 2100 needed that ratio near 2 (median 2.08, range 1.21 to 3.44).

ILLUSTRATION: simple no-feedback lines

How strong must it be, and by when?

Source: paper p.20-21 (Eq. 63-65), p.30, p.41. Lines drawn for the no-feedback case, a(t) = e^((P-1)t).

To tie by 2100, robots must add 3.0% a year to their own number. To tie by 2050, 26.9%.

That is about 9 times harder, with neutral feedback. It is on top of what factories already add.

SCENARIO: a threshold, not a forecast

Which guess matters most?

Sources: paper Table 3 p.28 (2100), Table 2 p.26 (2050, start 2035, no feedback). Computed: 0.269 / 0.030 = 9.0.

How long a robot lasts matters most.

The authors mixed up 4,096 sets of guesses. Longer-lived robots need fewer replacements, so the needed build rate falls.

Today's exact robot count barely mattered.

SCENARIO TEST: not a probability

So what should you watch for?

Source: paper Table S1 p.43, S1-S5 p.39-43. All links are negative: more of the input, lower the need.

Next time you see a robot headline, ask three things.

  • How many exist today? Counted, or guessed?
  • Who builds the next one? A factory, or a robot?
  • How long does one last before it is replaced?

Where could this math be wrong?

Source: paper p.37 (the authors call for counts of stock, gross production, retirement and robot-made share).

The upper body of a humanoid robot with a round dark face screen and a blue torso marked Boston Dynamics, seen from the front
Boston Dynamics Atlas, front view. Photo: Damian B Oh / Wikimedia Commons, CC BY-SA 4.0.

This is a what-if, not a forecast.

The robot counts are scenario numbers. The robots-building-robots strength was not measured at all, and rates are held constant. A tie means only a head count, not power, control or economic weight.

It is a preprint, and we found no peer review. It had 0 citations on 4 November 2026. The authors say ChatGPT Plus helped polish the text and check their math.

Want to check the sources yourself?

Sources: paper p.7, p.32, p.35-36, p.38; OpenAlex record, checked 4 Nov 2026.

Close view of a humanoid robot's round head and blue upper body, with Boston Dynamics printed on the chest
Boston Dynamics Atlas, close view. Photo: Damian B Oh / Wikimedia Commons, CC BY-SA 4.0.

Sources & method

  1. Zaccone A, Tunstall T, Ausubel JH. When will androids outnumber humans? arXiv:2611.03242v1, 2 Nov 2026. arxiv.org/abs/2611.03242
  2. UN DESA Population Division. World Population Prospects 2024: Summary of Results. desapublications.un.org
  3. Morgan Stanley Research, Humanoids: a $5 trillion market, May 2025. morganstanley.com
  4. Chamie J. The rise of androids among human populations. Inter Press Service, Sep 2025. ipsnews.net
  5. UN peak of about 10.3 billion in the mid-2080s: Dawn, 13 Jul 2024; IDN, 11 Jul 2024
  6. OpenAlex record for the paper (0 citations, checked 4 Nov 2026). api.openalex.org
  7. Computed here, inputs from the paper: 823,000 = 8.23e9 / 1e4 (p.8, p.10). 73.5% = e^(ln(9.664e9 / 1e4) / 25) - 1 (p.25). 0.388 = 3.96 / 10.19 (p.23). About 20x = 9.8% / 0.49% (p.9, p.11). About 9x = 0.269 / 0.030 (p.26, p.28).
  8. Not used: the paper's vehicle count (1.70 billion) and data-centre power figures. We could not check them.

Credit

The research is by Alessio Zaccone (University of Milan), Thomas Tunstall (University of Exeter) and Jesse H. Ausubel (The Rockefeller University). They build on Joseph Chamie's essays on androids as a population.

Story by vizmaya desk. More stories at vizmaya.fyi/stories.

Photos

  1. TOCABI black 3200px. Cadop, Wikimedia Commons, CC BY-SA 4.0
  2. Atlas frontview 2013. DARPA, Wikimedia Commons, public domain
  3. Ameca Generation 1. Willy Jackson, Wikimedia Commons, CC BY-SA 4.0
  4. Boston Dynamics Atlas (3). Damian B Oh, Wikimedia Commons, CC BY-SA 4.0
  5. Boston Dynamics Atlas (4). Damian B Oh, Wikimedia Commons, CC BY-SA 4.0